Anthropic Ships Claude Opus 4.7 With Cyber Throttle
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Anthropic released Claude Opus 4.7 to general availability on April 16, 2026, framing it as a "notable improvement" over Opus 4.6 in advanced software engineering and adding a new "xhigh" effort level
- The New Stack reports the model brings better vision, memory, and instruction-following, while VentureBeat says it narrowly retook the lead for most powerful generally available LLM
- Multiple outlets — including Neowin, iClarified, and Crypto Briefing — note Opus 4.7 beat OpenAI's GPT-5.4 on agentic coding benchmarks
- The Decoder and Implicator.ai highlight that Anthropic deliberately scaled back cyber capabilities, shipping the model below the unreleased "Mythos" safety tier
- Amazon Web Services added Opus 4.7 to Amazon Bedrock on launch day, and Sherwood News flags occasional "doom loops" alongside the coding and vision gains
- Decrypt's review calls the model a "token-eating machine," and Simon Willison found a smaller Qwen model beat Opus 4.7 on a pelican-drawing test
Why it matters: Anthropic is reclaiming the top agentic-coding benchmark while deliberately throttling cybersecurity capabilities to ship below the "Mythos" safety tier — a calibrated compromise that lets it monetize Opus 4.7 through AWS Bedrock on day one without reigniting the safety debate that stalled Mythos.
